130+ curated examples · copy and generate
Stickers, product loops and pixel animations. Pick one and generate it in Raphael.
Roaring duck
@ailingqu
Pixel knight run cycle
@ailingqu
Confused shiba head tilt
@ailingqu
Cat typing furiously
@ailingqu
Pixel art campfire
@ailingqu
Pelican riding a bicycle
@SSHCodes
Dwarf game model rotating 360 degrees
@xikhar
Penguin facepalm
@ailingqu
White bunny reaction sticker set
@ailingqu
Sneaker colorway flash
@ailingqu
@fal
Follow the red markup on this photo. Each number marks a spot and the legend box explains what to do there. Apply every instruction, then remove all of the markup, including the numbers and the legend box, so the result is a clean photo.
@OpenAI
In this room photo, replace ONLY the white chairs with chairs made of wood. Preserve camera angle, room lighting, floor shadows, and surrounding objects. Keep all other aspects of the image unchanged. Photorealistic contact shadows and fabric texture.
130 GPT Image 2.6 prompts with their real outputs, collected from OpenAI's documentation, launch materials and community creators, and synced from the open-source awesome-gpt-image-2-6 repository.
Cases come from the authors' published original prompts and are localized into your language while keeping in-image text unchanged — never descriptions reverse-engineered from images. Cards mark whether a case is official or from the community, and each detail view lets you inspect the original prompt and source.
Copy a prompt and swap in your own subject and style. For cases marked "Needs an image", add your own image first. The generator at the top of this page uses GPT Image 2.6 Sunburst by default.
Try the generator →| Version | Best for |
|---|---|
| GPT Image 2.6 Sunburst | Detailed creation and photo editing, such as adjusting objects, lighting and scene details. The generator on this page uses it by default. |
| GPT Image 2.6 Flare | Everyday creation and quick drafts, such as portraits, explainer diagrams and presentation visuals; it can also edit from reference images. |
Both support text-to-image and image-to-image, up to 4K, five quality levels, three planning levels and up to 16 reference images.
Before editing a portrait or product shot, decide what must not change, such as facial features, the product's shape or the text on its packaging, and say so in the prompt. Start detail-heavy edits with Sunburst.
Put the exact words in quotes and say where they go and in what style. Shorter text is easier to get right. Both versions can render text; still proofread every word before publishing.
Click "Try now" on any GIF case to load the prompt straight into the generator above in GIF mode (2 to 16 frames at 0.2 seconds each); for vector files, use the AI SVG Generator to output SVG directly.
Actual results and generation time depend on the request. The generate button shows the credits a request needs before you run it.
Common questions about GPT Image 2.6 prompts and generation
Yes. Cases come from OpenAI's documentation, launch materials and community creators, curated in the open-source awesome-gpt-image-2-6 repository. Prompts are localized into your language while keeping in-image text intact, and every detail view lets you toggle the author's original prompt and open its source link.
The same prompt gives a different result every time, and the author may have used other quality settings or reference images. Generate a few times, or swap in your own subject and style. For multi-step cases, follow the steps in order and use each result as the reference for the next step.
These prompts were written around reference images, for example placing the person from a photo into a poster. Upload your own images in the generator first (up to 16), then paste the prompt. The detail view shows the references the author used.
Click "Try now" on any case in GIF Lab or any GIF card on this page — the prompt loads straight into the generator at the top in GIF mode, where you can pick 2 to 16 frames (0.2 seconds each) and generate, or start from scratch in the AI GIF Generator.
Put the words in quotes, keep them short, and say where they go and in what font style; avoid too many text blocks in one image. Proofread the result, and fix a single wrong word with a local edit.
It depends on quality, size and the number of reference images, and for GIFs on the frame count. The generate button shows the exact credits before you run it.
Use an example prompt or enter your own idea to generate images with GPT Image 2.6 in Raphael.
Start generating →